Development and Application of Computer Vision in High-Intensity Interval Training

With the rapid development of deep learning, an increasing number of researchers are focusing on utilizing computer vision to enhance sports performance. This study proposes a novel framework to improve the safety and effectiveness of high-intensity interval training (HIIT) through real-time action recognition and fatigue monitoring. Our framework consists of a three-step pipeline: First, HIIT videos are fed into a human pose estimation model to extract the skeletal locations of keypoints in each frame. Next, a machinelearning-based classifier is employed to identify the specific action being performed by the participant. Finally, kinematic changes are monitored based on principles of sports biomechanics to determine whether the participant is experiencing fatigue. To ensure stability and reliability, a sliding window method is employed, and optimal hyperparameters are determined through experimentation. A HIIT dataset was developed for training and evaluation. Experimental results shows that the framework achieves satisfactory performance in both action recognition and fatigue monitoring.